Modeling Heat Transfer in an Urban Settlement with 3D Cellular Automata and Artificial Intelligence

This chapter presents the development and preliminary results of a three-dimensional (3D) cellular automata (CA) model designed to simulate urban heat transfer in an artifi-cial 50 × 50 × 20 m city block located at moderate latitude (Fhi = 46.24◦ N). Each model cell is assigned to one of five material classes: outdoor air, asphalt street, park/grass, structural concrete, and interior air and advances temperature states via eight sequen-tially applied physics layers: conductive diffusion, solar heating, nocturnal radiative cooling, free-atmosphere relaxation, asphalt boundary-layer plume convection, surface air Newton cooling, interior ventilation, and boundary conditions. A key aspect of this work is the workflow by which the simulation code was produced: rather than conven-tional manual coding, specified intent and domain constraints in natural language were set by human while Claude AI (Anthropic) ([27]) generated, debugged, and iteratively refined the Python implementation across multiple turns of dialogue. The human sup-plied physical intuition, validated outputs against expected temperature climate ranges, and performed targeted corrections where the AI-generated code produced physically implausible results. Over and underestimates were corrected through a joint debugging session. Preliminary simulation results over a 30-day May period demonstrate realistic diurnal temperature cycles, street-surface temperatures 4–6 ◦C above park surfaces at solar noon, and an asphalt boundary-layer plume decaying exponentially with height up to 10 m above ground. These results confirm the suitability of the CA paradigm for urban heat studies [5, 2] and illustrate the potential of large language model (LLM)-assisted research and rapid prototyping of physically complex simulations.